Self-Curable Synaptic Ferroelectric FET Arrays for Neuromorphic Convolutional Neural Network.

Area of Science:

• Materials Science
• Computer Engineering
• Neuroscience

Background:

• Deep learning advancements drive interest in neuromorphic computing, mimicking the human brain.
• Energy-efficient and reliable artificial synapses are crucial for neuromorphic systems.

Purpose of the Study:

• To fabricate and demonstrate a synaptic ferroelectric field-effect-transistor (FeFET) array for neuromorphic convolutional neural networks.
• To develop and validate an efficient self-curing method to enhance FeFET array endurance.

Main Methods:

• Fabrication of a synaptic FeFET array.
• Demonstration of synaptic weight potentiation/depression and program-inhibiting operations at the array level.
• Implementation of a self-curing method utilizing punch-through current and low-frequency noise spectroscopy for evaluation.

Main Results:

• Achieved array-level long-term potentiation and depression of synaptic weights.
• Demonstrated successful program-inhibiting operations with 79.84% learning accuracy on the CIFAR-10 dataset.
• Improved FeFET array endurance by tenfold using the proposed self-curing method.

Conclusions:

• The study provides a viable method for fabricating and operating reliable synaptic FeFET arrays.
• The developed self-curing technique significantly enhances device endurance, crucial for practical neuromorphic applications.
• This work facilitates the advancement of ferroelectric-based neuromorphic computing.